Atticus Li — Better Decisions

Behavioral Economy Every product you've shipped is already a behavioral intervention — the only open question is whether anyone is governing it.

Behavioral economics is not a list of biases — it's the discipline of studying how people actually decide, and most companies that claim to "use" it have never tested whether it works on their own users.

Loss Aversion Losses feel roughly two to two-and-a-half times more intense than equivalent gains — and that asymmetry is quietly deciding which growth leaders get hired, funded, and fired, long before anyone looks at the underlying data.

The Action-Velocity Signal: Why Raw Token Counts Lie About AI Agent Costs Ask most people running AI coding agents daily what their biggest cost driver is, and they'll point at a token counter.

The Three-Layer Audit: How to Actually Verify an AI Coding Agent Before You Trust It Most people treat an AI coding agent's output the way they treat a junior analyst's first draft: skim it, and if it sounds coherent, ship it.

Anthropic's own team gave a "Claude Code best practices" talk.

The prompt is not the skill.

Most cognitive bias lists are inventories.

*The foot-in-the-door technique is the most quietly over-prescribed tactic in conversion optimization — and its failure mode is nearly invisible on a dashboard.* Get someone to say a small yes, and the next, larger yes comes easier.

In November 2023, Andrej Karpathy — co-founder of OpenAI, former head of AI at Tesla — gave a one-hour public talk explaining how large language models actually work, no machine-learning background required.

A prospect forms their first impressions of your SaaS landing page in about five seconds before deciding whether to keep reading or leave.

Modern AI tools can function as a powerful landing page copy generator, writing 20 variants before your next meeting.

You ran the test.

A test can win and still lose money.

Before I run any experiment, I calculate the required sample size.

The Decision That Determines Everything Else Before you launch an A/B test, you make one decision that shapes the entire experiment: the minimum detectable effect.

The same experiment result can get a growth program funded for another year or gutted at the next planning cycle — and which one happens often has nothing to do with the number itself.

Repeat visitor tests can lie with a straight face.

It's a seductive idea.

The honest answer to "which AI coding tool is best" is annoying: it depends on the task.

"One-shot" has become a brag: paste a prompt, hit enter, get a working feature, screenshot it.

If every test feels urgent, you do not have a broken experimentation strategy.

There are two ways to cut your Claude Code token bill.

Inside the Trade-Off: How Dark Patterns Actually Get Designed Meta description: Regulators' complaints increasingly include the internal emails and test data behind a dark pattern, not just the pattern itself.

The A/B Test That Became a Regulatory Case Study Meta description: Credit Karma's "pre-approved" claim wasn't a lie — it was an A/B test winner, shipped the way any growth team ships a winner.

The Friction Was Never Symmetric Meta description: A good UX instinct — don't interrupt the flow of the experience — produces a very different outcome when the action being smoothed is a real-money purchase instead of a game action.

Inside the "Iliad" Flow: What Amazon's Cancellation Design Reveals About Retention Metrics vs.

One Click to Accept, Five to Refuse Meta description: A compliant-looking cookie banner and a compliant cookie banner aren't the same thing — the difference is measured in click counts and visual weight, not whether a banner exists at all.

No Fine, Just a Deadline Meta description: Six UK hotel-booking sites — and 25 more that followed — resolved a CMA consumer-protection investigation without paying a single pound in penalties.

The test report says the winning variant lifted conversion by a clean, quotable number.

Ask three tools in the same company how many conversions happened last month and you'll get three different answers.

Add up what every team in a large company claims to have driven, and the total will exceed the company's actual revenue — sometimes by a lot.

The experimentation playbooks are written by companies that can ship a test in an afternoon.

Every mature experimentation program is sitting on a graveyard of half-dead feature flags — tests that concluded months or years ago, whose code still ships to production, whose owners have left the company, and which nobody is confident is safe to delete.

Here is a result that should be impossible but isn't: a variant that wins on the topline, ships, and then underperforms — because it was actually worse for every single segment of users, and only looked better because the traffic mix between the two arms…

Every experimentation platform demo shows you the same thing: a clean setup wizard, a tidy results dashboard, a statistical engine that "just works."

The experimentation literature obsesses over getting to a valid result.

Some SaaS changes should raise revenue.

A channel's dashboard can go up on the same day total company revenue stands still — because the channel didn't create demand, it just caught demand that was headed there anyway.

Most low-traffic SaaS teams do not have a testing problem.

Which AI Model Should You Actually Use to Build Software?

When you are trying to spot bot traffic in A/B tests, the numbers can be alarming.

I have seen six-figure decisions ride on an event that never fired.

A test can win on paper and still lose money.

A winning test can still lose you money.

Most bad reruns do not fail in the stats tool.

If the same visitor sees variant A on Monday and variant B on Wednesday, your A/B testing efforts are not measuring behavior.

I have seen teams ship the wrong variant because week three landed on quarter end, and buyers stopped moving.

Most teams pick an A/B testing method based on who can ship faster this sprint.

You ran the test.